Last updated: 2026-05-21
Whitepapers & Research
Bayesian & Priors publishes research on AI agent channel measurement, the mechanics of the DSCRI-ARGDW pipeline, and the integration of gate scores as informative priors in Marketing Mix Models. Short research notes are available now. Full-length whitepapers are forthcoming in 2026.
Research program directed by Andres Plashal, founder of Bayesian & Priors and marketing scientist at Plashal LLC specializing in Bayesian econometrics and AI agent measurement.
01Research program
The research program at Bayesian & Priors covers three intersecting areas: the mechanics of the AAO framework (how gates interact), the econometrics of Bayesian MMM with a dedicated AI channel (how priors update), and the empirical landscape of AI-mediated recommendation (how, where, and when AI engines actually cite). Each output is published with a methodology section, reproducible math, and full citation back to source material.
02Research notes available now
-
Available
Multiplicative Gate Confidence: Why Ten Gates at 90 Percent Yield 35 Percent
End-to-end recommendation probability is the product of individual gate probabilities, not their average. This note derives the multiplicative model, demonstrates why a single weak gate dominates the result, and explains the bottleneck-first remediation strategy used by Bayesian & Priors. The full derivation lives in the AAO framework page. The composite rating reported to brands is the AAO Score.
-
Available
Prior, Likelihood, Posterior: The Three Components of a Bayesian MMM
A short primer on the Bayesian decomposition: how the prior encodes existing knowledge, how the likelihood reflects observed data, and how the posterior combines them. Includes the practical case for informative priors when measuring a previously unmeasured channel. The methodology lives in the MMM theory page.
-
Available
Why Traditional MMM Treats AI as Baseline Noise
Traditional MMM either lumps AI-mediated traffic into “organic” baseline (where it is indistinguishable from direct traffic) or applies uninformative priors that require 12 or more months of conversion data to stabilize. Either path produces lagged, low-confidence estimates that fail at allocation decisions.
03Forthcoming whitepapers
-
Forthcoming
Informative Priors from DSCRI-ARGDW Gate Scores: A Methodology
How gate scores translate into prior distributions on the AI channel coefficient in a Bayesian Marketing Mix Model. Covers prior calibration from sparse data, uncertainty propagation across gates, hyperparameter selection, and convergence behavior in the first 4 to 12 weeks of model life. Includes reference implementation in PyMC.
-
Forthcoming
The Corroboration Threshold: When AI Knowledge Treats Claims as Facts
Empirical analysis of the citation tipping point. At approximately 2 to 3 independent high-confidence sources, AI knowledge systems shift content from “claims to be” to “is.” Once crossed, the effect is multiplicative; early movers compound exponentially relative to competitors who have not. Includes measurement methodology and observed thresholds across ChatGPT, Claude, Perplexity, and Gemini.
-
Forthcoming
MMM Convergence Benchmarks: Informative vs Uninformative Priors on the AI Channel
Comparative analysis of convergence time, credible interval width, and allocation stability for Bayesian MMMs that use DSCRI-ARGDW priors versus standard uninformative priors. Reports benchmarks across simulated and real-world datasets at 6, 12, 26, and 52 weeks of model life.
-
Forthcoming
Sequential Bayesian Updating for Weekly DSCRI-ARGDW Re-scoring
Operational specification for how Bayesian & Priors blends each modeling period’s posterior with re-audited gate scores to produce the next period’s prior. Covers update frequency, blending weights, drift detection, and the conditions under which a hard prior reset is warranted.
04Frequently asked questions
How do I get the whitepapers when they publish?
Join the waitlist. Whitepapers are shared with waitlist members first, with a comment window for measurement leaders before public release. Public publication follows roughly 60 days after the comment window closes.
Can I cite Bayesian & Priors research?
Yes. Use the citation: Bayesian & Priors, a practice of Plashal LLC (baypri.ai). All research notes and whitepapers are published with a methodology section and explicit citation guidance.
What does the research program cover?
Three areas: the mechanics of the AAO framework (gate interactions, multiplicative confidence), the econometrics of Bayesian MMM with a dedicated AI channel (prior calibration, sequential updating), and the empirical landscape of AI-mediated recommendation (citation thresholds, platform weighting).